/* validate_msa:
* SRE, Thu Dec 3 16:10:31 2009 [J5/119; bug #h70 fix]
*
* HMMER uses a convention for missing data characters: they
* indicate that a sequence is a fragment. (See
* esl_msa_MarkFragments()).
*
* Because of the way these fragments will be handled in tracebacks,
* we reject any alignment that uses missing data characters in any
* other way.
*
* This validation step costs negligible time.
*/
static int
validate_msa(P7_BUILDER *bld, ESL_MSA *msa)
{
int idx;
int64_t apos;
for (idx = 0; idx < msa->nseq; idx++)
{
apos = 1;
while ( esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos]) && apos <= msa->alen) apos++;
while (! esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos]) && apos <= msa->alen) apos++;
while ( esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos]) && apos <= msa->alen) apos++;
if (apos != msa->alen+1) ESL_FAIL(eslEINVAL, bld->errbuf, "msa %s; sequence %s\nhas missing data chars (~) other than at fragment edges", msa->name, msa->sqname[idx]);
}
return eslOK;
}
/* Function: p7_Fastmodelmaker()
*
* Purpose: Heuristic model construction.
* Construct an HMM from an alignment by a simple rule,
* based on the fractional occupancy of each columns w/
* residues vs gaps. Any column w/ a fractional
* occupancy of $\geq$ <symfrac> is assigned as a MATCH column;
* for instance, if thresh = 0.5, columns w/ $\geq$ 50\%
* residues are assigned to match... roughly speaking.
*
* "Roughly speaking" because sequences may be weighted
* in the input <msa>, and because missing data symbols are
* ignored, in order to deal with sequence fragments.
*
* The <msa> must be in digital mode.
*
* If the caller wants to designate any sequences as
* fragments, it does so by converting all N-terminal and
* C-terminal flanking gap symbols to missing data symbols.
*
* NOTE: p7_Fastmodelmaker() will slightly revise the
* alignment if the assignment of columns implies
* DI and ID transitions.
*
* Returns the HMM in counts form (ready for applying Dirichlet
* priors as the next step). Also returns fake traceback
* for each training sequence.
*
* Models must have at least one node, so if the <msa> defined
* no consensus columns, a <eslENORESULT> error is returned.
*
* Args: msa - multiple sequence alignment
* symfrac - threshold for residue occupancy; >= assigns MATCH
* bld - holds information on regions requiring masking, optionally NULL -> no masking
* ret_hmm - RETURN: counts-form HMM
* opt_tr - optRETURN: array of tracebacks for aseq's
*
* Return: <eslOK> on success. ret_hmm and opt_tr allocated here,
* and must be free'd by the caller (FreeTrace(tr[i]), free(tr),
* FreeHMM(hmm)).
*
* Returns <eslENORESULT> if no consensus columns were annotated;
* in this case, <ret_hmm> and <opt_tr> are returned NULL.
*
* Throws: <eslEMEM> on allocation failure; <eslEINVAL> if the
* <msa> isn't in digital mode.
*/
int
p7_Fastmodelmaker(ESL_MSA *msa, float symfrac, P7_BUILDER *bld, P7_HMM **ret_hmm, P7_TRACE ***opt_tr)
{
int status; /* return status flag */
int *matassign = NULL; /* MAT state assignments if 1; 1..alen */
int idx; /* counter over sequences */
int apos; /* counter for aligned columns */
float r; /* weighted residue count */
float totwgt; /* weighted residue+gap count */
if (! (msa->flags & eslMSA_DIGITAL)) ESL_XEXCEPTION(eslEINVAL, "need digital MSA");
/* Allocations: matassign is 1..alen array of bit flags.
*/
ESL_ALLOC(matassign, sizeof(int) * (msa->alen+1));
/* Determine weighted sym freq in each column, set matassign[] accordingly.
*/
for (apos = 1; apos <= msa->alen; apos++)
{
r = totwgt = 0.;
for (idx = 0; idx < msa->nseq; idx++)
{
if (esl_abc_XIsResidue(msa->abc, msa->ax[idx][apos])) { r += msa->wgt[idx]; totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsGap(msa->abc, msa->ax[idx][apos])) { totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos])) continue;
}
if (r > 0. && r / totwgt >= symfrac) matassign[apos] = TRUE;
else matassign[apos] = FALSE;
}
/* Once we have matassign calculated, modelmakers behave
* the same; matassign2hmm() does this stuff (traceback construction,
* trace counting) and sets up ret_hmm and opt_tr.
*/
if ((status = matassign2hmm(msa, matassign, ret_hmm, opt_tr)) != eslOK) {
fprintf (stderr, "hmm construction error during trace counting\n");
goto ERROR;
}
free(matassign);
return eslOK;
ERROR:
if (matassign != NULL) free(matassign);
return status;
}
/* Function: p7_Alimask_MakeModel2AliMap()
* Synopsis: Compute map of coordinate in the alignment corresponding to each model position.
*
* Args: msa - The alignment for which the mapped model is to be computed. We assume
* the MSA has already been manipulated to account for model building
* flags (e.g. weighting).
* do_hand - TRUE when the model is to follow a hand-build RF line (which must be
* part of the file.
* symfraq - if weighted occupancy exceeds this value, include the column in the model.
* map - int array into which the map values will be stored. Calling function
* must allocate (msa->alen+1) ints.
*
* Returns: The number of mapped model positions.
*/
int
p7_Alimask_MakeModel2AliMap(ESL_MSA *msa, int do_hand, float symfrac, int *map )
{
int i = 0;
int apos, idx;
float r; /* weighted residue count */
float totwgt; /* weighted residue+gap count */
i = 0;
if ( do_hand ) {
if (msa->rf == NULL) p7_Fail("Model file does not contain an RF line, required for --hand.\n");
/* Watch for off-by-one. rf is [0..alen-1]*/
for (apos = 1; apos <= msa->alen; apos++) {
if (!esl_abc_CIsGap(msa->abc, msa->rf[apos-1]) ) {
map[i] = apos;
i++;
}
}
} else {
for (apos = 1; apos <= msa->alen; apos++)
{
r = totwgt = 0.;
for (idx = 0; idx < msa->nseq; idx++)
{
if (esl_abc_XIsResidue(msa->abc, msa->ax[idx][apos])) { r += msa->wgt[idx]; totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsGap(msa->abc, msa->ax[idx][apos])) { totwgt += msa->wgt[idx]; }
else if (esl_abc_XIsMissing(msa->abc, msa->ax[idx][apos])) continue;
}
if (r > 0. && r / totwgt >= symfrac) {
map[i] = apos;
i++;
}
}
}
return i;
}
